The Mess of Software Metrics
Capers Jones · Auerbach Publications eBooks · 2021
The software industry is one of the largest in the world. But software itself is not a tangible product and measuring application size has been a weak link for many years. The two oldest metrics for software have been “lines of code per month” (LOC) for development progress and “cost per defect” for bug repairs. Unfortunately, these metrics are flawed and fail to show true software economics. The LOC metric penalizes high-level languages and makes assembly look better than modern languages such as Objective C and Java. Also, for large systems coding is less than 30% of total cost, and the LOC metric is useless for requirements, architecture, design, and other non-code work. The cost per defect metric penalizes quality and is cheapest for the buggiest applications. A zero-defect software application would have an infinite cost per defect. Fortunately, the function point metric provides a useful alternative for measuring both software productivity and quality. Work hours per function point can be applied to requirements, design, code, and everything else. Defect repair costs per function point are much more accurate than cost per defect, because they go down as quality goes up.